Hello, It has been said that in the validation phase the image list will be augmented using the test set of the leaderboard phase. We now are assuming that the image ID's of the leaderboard phase remain the same so we do not mess all the images from our previous partitions. Is this correct? Thanks

Created by Alberto Albiol alalbiol
> so it means that we have to start again the preprocessing on a larger dataset with a shorter, 288 hours, server time allowance? The purpose of the Validation Phase is not for you to redo the work that you have done during the Leaderboard Phase (e.g. testing multiple pre-processing approaches). Here, you are expected to do 1x pre-processing, 1x training for SC1 and/or 1x training on SC2. From our data, we are expecting that 288 hours should be enough for that purpose. > Do you mind clarifying how many images are there in the training set? Approximatively 450k. Please use `/progress.txt` to easily track the progress rate of your pre-processing/training submissions. Thanks!
Hi Thomas, so it means that we have to start again the preprocessing on a larger dataset with a shorter, 288 hours, server time allowance? Do you mind clarifying how many images are there in the training set? I don't think I have found a clear description. Maybe I missed it somewhere. Thanks, Michael
Hi Alberto, > We now are assuming that the image ID's of the leaderboard phase remain the same so we do not mess all the images from our previous partitions. The content of all the `/preprocessedData` directories have been cleaned before the start of the Validation Phase (mentioned in the newsletter). Hope this helps!

about images ID in validation phase page is loading…